Episode 8.05: ChatGPT isn’t a program, doesn’t store facts in words, and makes mistakes. Explained!

We look at some of the ways in which the chatbot doesn’t operate like an ordinary program such as some of us would know how to write. It doesn’t store information in atomic units, such as the name of the 44th president of the United States. And it does make mistakes, even though it claims that it can’t. We explain all these things, and indeed, why it makes mistakes, and how to some extent they are unavoidable given the way in which a neural net operates.